Exploring Multiple High-Scoring Subspaces in Generative Flow Networks
Xuan Yu, Xu Wang, Rui Zhu, Yudong Zhang, Yang Wang · 2026
Read Exploring Multiple High-Scoring Subspaces in Generative Flow Networks by Xuan Yu, Xu Wang, Rui Zhu, Yudong Zhang, Yang Wang (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-11491.txt.
Description
Generative Flow Networks can over-explore vast spaces, sampling low-reward regions too often. Xuan Yu and four co-authors propose CMAB-GFN, which joins a combinatorial multi-armed bandit frame to flow network policies. The bandit prunes low-quality actions, yielding compact high-scoring subspaces for exploration.
Restricting exploration to those subspaces speeds discovery of high-value candidates, while exploring other subspaces preserves diversity. Tests on multiple tasks show higher-reward candidates than current approaches. The work addresses a core tension in flow networks: exploration must be broad enough to find good regions, yet focused enough to avoid wasting samples on poor ones.
Catalog record
| CREATOR | Xuan Yu, Xu Wang, Rui Zhu, Yudong Zhang, Yang Wang |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 64 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-11491